Efficient parallel generation of many-nucleon basis for large-scale ab initio nuclear structure calculations

0491174 - ÚJF 2019 RIV DE eng C - Konferenční příspěvek (zahraniční konf.)
Langr, D. - Dytrych, Tomáš - Oberhuber, T. - Knapp, F.
Efficient parallel generation of many-nucleon basis for large-scale ab initio nuclear structure calculations.
Lecture Notes in Computer Science. Berlin: Springer Verlag, 2018, s. 341-350. ISBN 978-3-319-78054-2. ISSN 0302-9743.
[12th Internatnional Conference on Parallel Processing and Aplied Mathematics (PPAM 2017). Czestochowa (PL), 10.09.2017-13.09.2017]
Grant CEP: GA ČR GA16-16772S
Institucionální podpora: RVO:61389005
Klíčová slova: Ab initio * basis generation * many-nucleon basis * nuclear structure * parallel algorithm
Obor OECD: Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

We address the problem of generating a many-nucleon basis for ab initio nuclear structure modeling, which quickly becomes a significant runtime bottleneck for large model spaces. We first analyze the original basis generation algorithm, which does not employ multi-threading parallel paradigm. Based on the analysis, we propose and empirically evaluate a new efficient scalable basis generation algorithm. We report a reduction of basis generation runtime by a factor of 42 on the Blue Waters supercomputer and by two orders of magnitude on our test-bed computer system with Broadwell CPUs.
Trvalý link: http://hdl.handle.net/11104/0285219